Anthropic Launches Claude Fable 5.2 for Paid Plans, Showcases Complex Science and Hardware Design Feats
Anthropic has rolled out Claude Fable 5.2 to all paid plans. Discover the model's pre-release achievements, including routing 4-layer PCBs, deriving organic che
On November 9, 2026 (local time), Anthropic officially launched Claude Fable 5.2, making it immediately available to all paid plan subscribers. Simultaneously, the company opened applications for Claude Mythos 5.2, a specialized model restricted to cybersecurity and biomedical research organizations.

Image source: https://x.com/claudeai/status/2119899459761131615 (@claudeai)
The launch is heavily anchored in the models' pre-release testing results, which demonstrated complex problem-solving capabilities across scientific research and engineering domains.
End-to-End Hardware Design and Manufacturing
In the hardware engineering space, Fable 5.2 successfully designed three functional electronic devices in under a day, operating entirely from a single-paragraph prompt.
The model independently wrote the specifications, selected components, routed 4-layer printed circuit boards (PCBs), designed 3D printable enclosures, and programmed the underlying software. Anthropic reported that all three physical prototypes—including a home voice assistant hub—worked without any design modifications.
Breakthroughs in Organic Chemistry and Theoretical Physics
Academic milestones were similarly notable. Using only coding tools and no external internet access, Fable 5.2 formulated complete chemical synthesis routes for four complex molecules in approximately 30 minutes. Researchers from Stanford, Yale, and Caltech rated the output at an "early graduate student" level. While these chemical routes have yet to be physically tested in a wet lab, automating molecular synthesis planning marks a significant computational step.
In theoretical physics, Fable 5.2 completed an intense calculation regarding black hole formation that had remained unsolved for 23 years, and identified the exact limits of a 15-year-old technique known as "modular bootstrapping." Both findings have been released as preprint papers.
Setting New Baselines in MIPLIB Optimization
Anthropic also published Fable 5.2's results against MIPLIB, a library of mathematical optimization problems. Out of 203 open problems, the model beat the listed mathematical solutions in 72 cases and successfully proved optimality in 45 cases. In one practical test—scheduling 42 trains—the AI agent generated a timetable that reduced peak power usage by 2.6% in under two hours, proving that no better timetable exists.
The fact that Fable 5.2 can independently verify chemical synthesis routes using only code tools, optimize complex logistical routing, and design a fully functional 4-layer PCB from scratch demonstrates a critical shift. It shows modern AI evolving beyond language-based knowledge retrieval into an applied "engineering problem-solver" capable of addressing real-world physical and logistical constraints.
Organizations looking to deploy the high-stakes Mythos 5.2 for cybersecurity and biotech applications must undergo a formal application process, as it will not be generally available.
Sources
- Anthropic Official Blog: Claude Fable and Mythos 5.2
- Claude Official X (@claudeai): 2026-11-09 Fable 5.2 Announcement Thread